英文摘要 | Image segmentation is one of the core issues in computer vision, and is the basis for image analysis and image understanding. The segmentation results of traditional data-driven segmentation methods, due to their own limitations, are difficult to meet the needs of the complex image segmentation applications, such as medical image segmentation. Therefore, there is an urgent need for a framework which combines the low-level visual characteristics of images and human priori knowledge, so that the segmentation results can be more accurate and selective. In this demand, the active contour models come into being. In recent years, active contour methods have become a hot research topic in image segmentation. The good performances of active contour model make it widely used in many areas, such as medical image analysis, remote sensing image processing, video tracking. In this context, this paper makes a research on the theories and methods of active contour models for image segmentation, and presents several effective algorithms in the aspects of intensity inhomogeneous images segmentation, the combination of edge-based and region-based active contour, and the global minimum of the Chan-Vese model. The main work in this thesis can be summarized as follows: 1) Considering the existing methods cannot segment intensity inhomogeneous images, we propose a Neighborhood Image Fitting (NIF) model which utilizes the neighborhood information. The NIF model fits the areas of the image by calculating the similarity between the central pixel and its neighbors. NIF model utilizes local area information to approximate the original image, and more details can be obtained. Therefore, NIF model can segment intensity inhomogeneous images. Experimental results show that, compared to the Chan-Vese model, the NIF model can correctly segment intensity inhomogeneous images; compared to the LBF model, the accuracy of the NIF model is higher. 2) Considering the pros and cons of the edge-based and region-based active contour models, we propose a Hybrid Geodesic Region-based Model(HGRM) for image segmentation. The energy term of HGRM model consists of three parts: edge term, global region term and local region term. Dut to the combination of edge-based and region-based active contour models, the HGRM model can segment noisy images and inhomogeneous images correctly. Experiments demonstrate that the model proposed can segment noisy images and weak boundary images with higher accuracy,... |
修改评论